Randomness as the Termination Mechanism for N≥3 Strategy Games

In a strategy game with N ≥ 3 players, full attack interaction, and public state, the deterministic equilibrium is gang-up-on-the-leader — coalitions form to suppress whoever’s ahead, compressing everyone back to parity, forever. The game theoretically never ends. Randomness is what lets it terminate at all — the kill switch on the dogpile equilibrium, not a flaw. So “random elements ruined the strategy” usually misreads what the randomness is for.

Links: The Multiplayer Coalition Problem (the gang-up equilibrium this terminates), Risk — The Attrition Constant (a second stabilizer in the same slot — Risk’s exact combat math shows the brake is not randomness per se but the cost of a won battle: attacking is tactically free of deterrence, yet the winner retains only ~15% of its force, so aggression is punished by the bystander rather than by the defender. Design lesson in the same family: omit tactical deterrence, recover stability from N), Bilateral Trade Valuation, Game Theory as Normative, Not Descriptive, Battleship — 30 Billion Boards (the N=2 contrast case — zero-sum, so a clean computable minimax equilibrium does exist, exactly the regime that breaks once N≥3)

A thesis page (portable, source-independent), promoted from a working note. NA1 is its canonical worked specimen; Catan and Diplomacy are its confirming/contrasting cases.

The theorem

For any game with (1) N ≥ 3 players, (2) full attack/interference interaction (no diplomatic immunity), (3) public state (a visible leader), and (4) no randomization, rational play is:

identify the visible leader → form a temporary coalition to suppress them → compress back to parity → dissolve the coalition → repeat.

Result: no terminal state. The game oscillates indefinitely around mid-tier parity. (The companion result is rubber-banding: the gang-up is a rubber band; this page is its design-side corollary — what you must add so the band doesn’t trap the game forever.)

Why randomness specifically

Strategy games could terminate other ways; each warps the game differently. Randomization is the one that doesn’t:

Mechanism	Examples	Trade-off
Time / turn limit	Civ turn cap, tournament chess	imposes a deadline that warps the strategic horizon; favors short-term plays
Resource exhaustion	Tigris & Euphrates, Power Grid	works, but only for resource-based games
Score threshold	most Eurogames	strategies aligned with the scoring axis dominate
Triggered condition	Civ space race, Risk domination	requires specific build-up paths; narrows strategic diversity
Randomization	NA1, Catan, Risk, Diplomacy (barely)	terminates probabilistically; strategy shifts the probability, not the certainty

Randomization uniquely keeps the strategic horizon open (no deadline), favors no single axis (unlike score caps), preserves agency (better doctrine → higher EV), accommodates very different game lengths (NA1’s 22 years vs Catan’s 10 turns), and lets asymmetric starts win sometimes (high-variance positions keep non-zero outcome probability). That makes it the most elegant termination mechanism for games where the strategic depth is the point.

Calibration is the design quality

The amount is a dial, not a toggle:

Too little → the game stalls (chess-like draws scale badly to N≥3).
Too much → pure luck; strategy stops mattering.
Sweet spot → random enough to terminate, deterministic enough that strategy dominates over many plays.

So the valid critique is never “randomness shouldn’t exist” but “randomness is miscalibrated for this game’s strategic depth.” And strategy quality is measured statistically over many plays, not by a single outcome — a great doctrine maximizes EV, not win-probability in any one game.

Evidence ledger (specimens)

Nobunaga’s Ambition (NA1) — the canonical well-calibrated specimen. Six independent randomization sources (daimyo-health/assassin RNG, bribe-defection counts, harvest bonus, combat skill+luck, event spawns, turn-order shuffle); remove any one and others still terminate the game, remove all and it stalls. The historical Honnō-ji assassination is the mechanic’s real-world instance — Oda was clearly leading and a coalition was forming when the assassination ended his run first. (See the NA1 crucible index — NA1’s in-vault hub; the deep RE artifacts live in the na1-decompiler repo, an external-repo specimen per the two-altitude method.)
Catan — dice + dev cards break the resource-scarcity stalemate and let weaker players score wins; the randomness is what keeps the N≥3 board from settling into perpetual leader-suppression.
Diplomacy — the contrast case that proves the rule: famously prone to multi-hour stalemates because it has almost no randomization layer. “Draws are the default” is community folk wisdom; Gunboat/time-pressure variants add the missing termination pressure.

Implications